Video summary

Generative AI vs AI agents vs Agentic AI

Main summary

Key takeaways

Technology

Technological Concepts: Generative AI vs AI Agents vs Agentic AI

1) Generative AI (LLM-based content generation)

  • Purpose: Generate new content (text, images, videos) by learning patterns from large datasets.
  • Core technology: Large Language Models (LLMs) such as GPT-4, Claude, Gemini, etc.
  • Training/data: Trained on large-scale internet knowledge (e.g., Wikipedia-like text, Google Books).
  • Limitation: Usually has a knowledge cutoff, so it can’t reliably answer time-sensitive questions (e.g., “price of a flight ticket tomorrow”).
  • Getting freshness (solutions):
    • If the LLM is allowed to search the web, it can return latest information.
    • It can call external APIs (e.g., Xedia, MakeMyTrip, travel APIs) when granted access to fetch current data.

2) AI Agents (tool-using, action-capable systems for tasks)

  • Analogy: An LLM is like a “brain”; tools/APIs are like “hammers and screwdrivers.”
  • What changes vs pure generative AI:
    • The system doesn’t just produce an answer—it can take actions.
    • It can use tools (APIs) and perform decisions to complete tasks.
  • Example (flight booking):
    • User asks: “Book the cheapest flight tomorrow from A to B.”
    • The agent uses travel APIs to:
      1. Search multiple options
      2. Select the cheapest
      3. Book the flight
  • Key characteristics:
    • Autonomous decision-making within a defined scope.
    • Task completion beyond Q&A: can plan steps, use tools, and execute actions.
  • Limitation noted: Early examples are often narrow; more complexity requires richer planning.

3) Agentic AI (multi-step, multi-agent, long/complex autonomous workflows)

  • Definition: Systems where one or more AI agents work autonomously—often for long and complex tasks—making decisions using tools, memory, and sometimes other agents.
  • Expansion example (complex trip planning):
    • Criteria include:
      • Destination: New Delhi
      • Month: May
      • Weather must be sunny for all 7 days
      • Budget under $1600
      • No layovers
    • Agentic behavior:
      • Calls a weather API to find 7 consecutive sunny days
      • Searches flights, compares options, and filters by budget/no layovers
      • Recommends hotels and airport taxis
  • Further expansion (multi-agent orchestration):
    • Adds an immigration/visa agent
    • Visa agent uses:
      • immigration APIs
      • user documents stored in a location (example: OneDrive with passport/records)
    • The system performs multi-step planning, such as checking visa eligibility before booking flights.
  • Control requirement: Not fully autonomous/safe—you still need human/system control (e.g., don’t allow agents to receive sensitive credentials like bank passwords).

  • Tooling/building agentic AI:

    • Mentions n8n with a workflow diagram where an LLM (e.g., Gemini) is a core component inside the agentic pipeline.
    • Notes multiple frameworks exist (e.g., Agno) with differing definitions.

    • Mentions that the creator of “Agno” defines agentic systems in five levels, but the video emphasizes the core progression:

      Generative AI → AI agent → Agentic AI

    • This progression corresponds to increasing task complexity, tool usage, and coordination/planning.

Reviews / Guides / Tutorials Called Out (Resources)

  • Llangraph tutorial: Explains how to build AI agents using the Llangraph framework.
    • Includes examples such as:
      • a chatbot with tools
      • memory
      • human-in-the-loop
  • AI boot camp project (more fully agentic system):
    • Example use-case: onboard an employee
    • Demonstrated capabilities:
      • Add employee to HRMS
      • Send welcome email
      • Notify the manager
    • Mentions architecture details:
      • Clot desktop as front end (spelled “clot” in subtitles)
      • MCP server as backend

Main Speakers / Sources

  • Speaker/source: The video narrator (creator) speaking directly in the subtitles.
  • Technologies/models referenced: ChatGPT (GPT-4), Claude, Gemini (LLMs).
  • Tool/API examples referenced: Xedia, MakeMyTrip, weather APIs (e.g., AccuWeather), immigration APIs, OneDrive.
  • Frameworks/tools referenced: Langraph, n8n, Agno.

Original video